Zinan Guo*, Yanze Wu*✝, Zhuowei Chen, Lang Chen, Peng Zhang, Qian He
(*Equal Contribution, ✝Corresponding Author)
ByteDance Inc
- 2024.10.31: 🔥 We are happy to release our latest models, PuLID-v1.1 and PuLID-FLUX-v0.9.1. See more in Model Zoo and pulid v1.1 model. We also update a new revision for the arXiv paper, which includes more results, details, and analysis, please check it out.
- 2024.09.26: 🎉 PuLID accepted by NeurIPS 2024
- 2024.09.12: We're thrilled to announce the release of the PuLID-FLUX-v0.9.0 model. Enjoy exploring its capabilities! 😊 Learn more about this model
- 2024.05.23: share the preview of our upcoming v1.1 model, please stay tuned
- 2024.05.01: release v1 codes&models, also the 🤗HuggingFace Demo
- 2024.04.25: release arXiv paper.
Please check the doc and demo of PuLID-FLUX here.
- Local gradio demo is ready now
- Online HuggingFace demo is ready now
- We have optimized the codes to support consumer-grade GPUS, and now PuLID-FLUX can run on a 16GB graphic card. Check the details here
- (Community Implementation) Online Replicate demo is ready now
- Local gradio demo supports 12GB graphic card now
- v0.9.1 is ready now
Below results are generated with PuLID-FLUX.
Images generated with our PuLID
Applications
pulid_demo.mp4
| Version | Base Model | Description |
|---|---|---|
| PuLID-v1 | SDXL | Paper model. |
| PuLID-v1.1 | SDXL | Compared to PuLID-v1, better compatibility, editability, facial naturalness, and similarity. |
| PuLID-FLUX-v0.9.0 | FLUX | Our first version for PuLID-FLUX, better prompt-following and image quality (since FLUX is more powerful than SDXL). But ID fidelity is not high enough for some male inputs |
| PuLID-FLUX-v0.9.1 | FLUX | Compared to PuLID-FLUX-v0.9.0, better ID fidelity. From the quantitative metric of ID similarity, the improvement is about 5 percentage points. Meanwhile, the editability remains similar as before. |
- Python >= 3.9 (Recommend to use Anaconda or Miniconda)
- PyTorch >= 2.0 if you don't need flux-dev-fp8, otherwise PyTorch >= 2.4.1
# clone PuLID repo
git clone https://github.com/ToTheBeginning/PuLID.git
cd PuLID
# create conda env
conda create --name pulid python=3.10
# activate env
conda activate pulid
# Install dependent packages
# 1. if you don't need flux-fp8, e.g., you are using xl or flux-bf16, install the following requirements.txt
pip install -r requirements.txt
# 2. if you need flux-fp8 (to put flux on consumer-grade gpu), install the following requirements_fp8.txt
pip install -r requirements_fp8.txt# for v1 version
python app.py
# for v1.1 version
python app_v1.1.py --base BASE_MODEL
Usage:
-base: can be RunDiffusion/Juggernaut-XL-v9 or Lykon/dreamshaper-xl-lightningThanks for the GPU grant from HuggingFace team, you can try PuLID HF demo in
- https://huggingface.co/spaces/yanze/PuLID for SDXL
- https://huggingface.co/spaces/yanze/PuLID-FLUX for FLUX
Following are some third-party implementations of PuLID we have found in the Internet. We appreciate the efforts of the respective developers for making PuLID accessible to a wider audience. If there are any PuLID based resources and applications that we have not mentioned here, please let us know, and we will include them in this list.
- Colab: https://github.com/camenduru/PuLID-jupyter provided by camenduru
- Replicate (PuLID): https://replicate.com/zsxkib/pulid provided by zsxkib
- Replicate (PuLID-FLUX): https://replicate.com/zsxkib/flux-pulid provided by zsxkib
- https://github.com/cubiq/PuLID_ComfyUI provided by cubiq, native ComfyUI implementation
- https://github.com/ZHO-ZHO-ZHO/ComfyUI-PuLID-ZHO provided by ZHO, diffusers-based implementation
- SD.Next Implementation provided by vladmandic
- Mikubill/sd-webui-controlnet#2838 provided by huchenlei
- PuLID-FLUX multi-person generation with Regional-Prompting-FLUX, provided by Anthony
This project strives to impact the domain of AI-driven image generation positively. Users are granted the freedom to create images using this tool, but they are expected to comply with local laws and utilize it responsibly. The developers do not assume any responsibility for potential misuse by users.
If PuLID is helpful, please help to ⭐ the repo.
If you find this project useful for your research, please consider citing our paper:
@InProceedings{guo2024pulid,
title={PuLID: Pure and Lightning ID Customization via Contrastive Alignment},
author={Guo, Zinan and Wu, Yanze and Chen, Zhuowei and Chen, Lang and Zhang, Peng and He, Qian},
booktitle={Advances in Neural Information Processing Systems},
year={2024}
}If you have any comments or questions, please open a new issue or feel free to contact Yanze Wu and Zinan Guo.